Deepfake In The Wild

BJ Zi, XJ Ma, MH Chang, JJ Chen, YG Jiang

Contact us

Abstact

In recent years, the abuse of a technique called Deepfake, which can be used to synthesize facial images, has drawn public attention. Recently, datasets such as FaceForensics++ and Deepfake Detection have been proposed to fill the gap in deepfake detection datasets. However, due to the poor quality of videos in these datasets, it is quite different from Deepfake videos on the Internet, especially Deepfake videos uploaded on the video sharing website, which hinders the further development of technologies to detect Deepfake videos. First，To facilitate the study of Deepfake detection in real-world situations, we proposed a dataset, Deepfake In The Wild. The videos in our dataset are completely collected on the Internet, consisting of 707 Deepfake videos and 7314 face sequences extracted from these videos. We annotated each face sequence as real or fake so that we could carry out Deepfake detection on face sequence level as well as image level. Second, we proposed a novel method to detect Deepfake videos. The method forge the attention mask used in Deepfake technology. And then, the forged attention mask is used as a attention to increase the weight of the infomation of face generated by Deepfake. Extensive demonstrate have shown that our network performs favorably in the Deepfake detection methods.

Introduction

Previous datasets

Dataset name	Download	Generate method	Deepfake videos	Actors
Deepfake-TIMIT low	download	Deepfake	320	32
Deepfake-TIMIT high	download	Deepfake	320	32
Faceforensics	-	Deepfake	1000	977
Faceforensics++	download	Deepfake	1000	977
Deepfake detection	download	Deepfake	over3000	28
Celeb-deepfakeforensics	download	Deepfake	795	13
DFDC	download	Deepfake	-	-

Ours

Dataset name	Download	Generate method	Deepfake videos	Actors
Deepfake in the wild	download	Download from Internet	707	about100

Examples from previous datasets

TODO

Deepfake-TIMIT low

Deepfake-TIMIT high

FaceForensics

None

FaceForensics++

Deepfake Detection

Celeb-deepfakeforensics

DFDC

examples from ours

Deepfake in the wild

File Structure:

deepfake_in_the_wild
                    |--real train
                                 |--0.tar.gz
                                 |--1.tar.gz
                                 |--2.tar.gz
                                 ...
                    |--real test
                                |--0.tar.gz
                                |--1.tar.gz
                                |--2.tar.gz
                                ...
                    |--fake train
                                 |--0.tar.gz
                                 |--1.tar.gz
                                 |--2.tar.gz
                                 ...
                    |--fake test
                                |--0.tar.gz
                                |--1.tar.gz
                                |--2.tar.gz
                                ...

In each tar.gz file, there will be several folders containing face images, and the images in each folder represent a face sequence. The image name in the folder represents the frame number it appears in the original video.

Highlights

-Collected from Internet

-well made

Our Method

TODO

We will upload our method in few days. Our method achieves state of the art in many datasets.

Expriments

First, we use pretrained Resnet-101 to extract features from the images in previous datasets and our dataset. Then we use the T-SNE to reduce the dimensionality. Red points represent fake faces, green points represent real faces. Here is the comparsion:

Download

Baidu Drive

passwd:izr7

Warning:

This is not our final vision.

knightofdawn / deepfake_in_the_wild Goto Github PK